by Passive data — the behavioral signals customers leave rather than the answers they give — is quietly reshaping consumer research. AI is usually sold into the field as a speed upgrade, running the same interviews and focus groups faster and cheaper, but its more useful effect has less to do with speed than with the kind of evidence research can rely on. Even the best-run qualitative methodology has always run into the same ceiling, and it is sharpest in business-to-business categories — the needs that matter most are the ones customers cannot or will not state on request. When the important information never makes it into an answer, asking better is not the fix.
The more productive move is to treat research less as elicitation and more as observation, and to let AI do what people cannot at scale — read those signals continuously, across a whole population rather than a recruited handful. The point is not that AI replaces the interview, but that it expands qualitative work beyond it.
The article is based on a session delivered by Simon Poon, Head Strategist, Ksher Global, presented at QUAL360 APAC 2025.
1. The real constraint in B2B research is not method — it is that customers cannot tell you what they need
In business-to-business categories, the most important needs are rarely the ones a customer will articulate when asked. A merchant running day-to-day operations has no direct channel to tell a supplier what is quietly costing them, and often cannot name it themselves; the friction is absorbed into business-as-usual until it surfaces as churn. Traditional qualitative research is built to ask well — structured guides, careful moderation, deliberate preparation — but it inherits a set of constraints that matter more as the questions get harder: small samples, a heavy dependence on the individual researcher’s skill, and a picture taken at a single point in time.
None of those constraints is a failure of rigor. They are the cost of a method organized around elicitation — around getting the right person to say the right thing in a scheduled conversation. The practical consequence is that a research function can run flawless sessions and still miss the needs its customers never put into words. That gap is the starting point for a different question: not how to ask better, but how much can be understood without asking at all.
2. Passive intelligence shifts the evidence from stated answers to observed behavior
Passive intelligence is the systematic collection and analysis of consumer behavior without direct questioning — reading what people do rather than what they report. Its defining properties are that it is unobtrusive, captures behavior in real-world contexts, runs continuously rather than as a snapshot, and operates at a scale traditional fieldwork cannot reach. The useful distinction is between active and passive digital footprints. Active footprints — posts, reviews, ratings, reactions — are high on intent and self-awareness, because the person knows they are being seen. Passive footprints — browsing paths, time spent on a page, location and device signals — score lower on intent but higher on authenticity and richness, precisely because they are not performed for an audience.
The method is triangulation across fragments rather than a single elicited answer. In a merchant context, that can mean linking location activity to points of interest to infer that a cluster of accounts in one district behaves like wholesale traders, or reading consistent patterns in review scores across a seller’s products as a signal of quality and business style. No single data point is decisive; the picture forms from many weak signals assembled into a stronger one. The trade is deliberate — a mix of active and passive footprints supports the segmentation and journey mapping that neither source delivers alone.
3. What AI removes is the ceiling on sample size and timing
Consumer research has moved in steps — from focus groups and surveys, to online panels and digital analytics, to social listening, and now to AI-run qualitative work — and each step loosened a different constraint. What AI removes is the pair that defined qual’s limits: the small sample and the single point in time. Analysis can run continuously, draw on the full population of available signals rather than a recruited subset, and track change over time instead of freezing a moment. Software agents extend the same logic to unstructured material, working asynchronously and to consistent criteria — pulling product mentions, sentiment, and audience reactions from video, or monitoring conversations across platforms at a volume no team could staff.
The interview itself changes shape rather than disappearing. An AI-run micro-interview embedded in a messaging thread can sustain roughly ten exchanges inside a five-minute conversation, adapting each question to the last answer. In the figures presented, that format drew about 142 percent more words than open text fields and reported user satisfaction near 93 percent — illustrative of a broader point, which is that short, frequent, conversational probes can gather richer material than a longer form people abandon.
4. The judgment stays human — and the standard for it rises
The design that holds this together is a loop, not a hand-off. Behavioral signals and AI-run conversations feed a common store; that raw material is then enriched through three layers working together — expert evaluation, statistical algorithms, and large language models — before it is applied to decisions about product, pricing, promotion, and channel, and back into both qualitative and quantitative research. Expert judgment is not the step AI replaces; it sits inside the enrichment layer, deciding what the signals are allowed to mean.
That matters because reading behavior without asking raises the interpretive bar rather than lowering it. A stated answer carries its own context; an inferred one does not, and the risk of reading intent into a pattern that has none grows with the scale of the data. The discipline that made traditional qualitative work credible — knowing which evidence to trust and how far to push it — becomes more important, not less, when the volume of evidence multiplies.
Conclusion
The operating model Simon Poon set out at QUAL360 APAC 2025 is less about adding AI tools than about changing what a B2B insight function treats as evidence — weighting observed behavior alongside stated preference, and building a loop in which passive signals and AI-run conversations sharpen each other while human judgment governs the reading. Understanding customers who cannot fully articulate their own needs was always the hard part of business-to-business research. The teams likely to pull ahead are not the ones that simply buy the tools, but the ones that redesign the loop and hold the line on interpretation.
Continue the conversation at QUAL360 APAC 2026
How much of what your customers really think never makes it into a research session?
QUAL360 is the only dedicated qualitative research conference in Asia, with 18 years of experience bringing together the region’s senior insights and research community. The 2026 edition in Singapore gathers leaders from Pandora, Symrise, Google, Great Eastern and more for two days of case studies, peer exchange, and practical discussion on where qualitative research is headed.
QUAL360 APAC 2026 | November 3–4 | Singapore. Explore the event.
